Seedream Image Skill

An AI Agent Skill for generating and editing images using ByteDance's Seedream models via Atlas Cloud API.

Works with 15+ AI coding agents including Claude Code, Cursor, OpenAI Codex, GitHub Copilot, Gemini CLI, Windsurf, OpenCode, Kiro, and more.

Seedream is ByteDance's state-of-the-art image generation model family, renowned for its exceptional Chinese text rendering, typography, and poster design capabilities. This skill brings Seedream's power directly into your AI coding workflow.

Features

Text-to-Image Generation — Generate stunning images from text prompts with industry-leading quality
Image Editing — Edit existing images with text instructions, masks, and strength control
Batch / Sequential Generation — Generate up to 15 related images in a single API call
Typography & Poster Design — Best-in-class Chinese character rendering and layout
NSFW-Capable — Whitelisted for unrestricted content generation on Atlas Cloud
Multiple Model Variants — Choose the right model for your use case and budget
Flexible Output — Control size, steps, guidance, seeds, and more

Model Variants

Model	ID	Starting Price per Image	Description
Seedream v5.0 Lite	`bytedance/seedream-v5.0-lite`	from $0.032/image	Latest generation, fast and high quality
Seedream v5.0 Lite Edit	`bytedance/seedream-v5.0-lite/edit`	from $0.032/image	Image editing with v5.0 Lite
Seedream v5.0 Lite Sequential	`bytedance/seedream-v5.0-lite/sequential`	from $0.032/image	Batch generation up to 15 images
Seedream v5.0 Lite Edit Sequential	`bytedance/seedream-v5.0-lite/edit-sequential`	from $0.032/image	Batch editing up to 15 images
Seedream v4.5	`bytedance/seedream-v4.5`	from $0.030/image	Previous generation, proven quality
Seedream v4.5 Edit	`bytedance/seedream-v4.5/edit`	from $0.030/image	Image editing with v4.5
Seedream v4.5 Sequential	`bytedance/seedream-v4.5/sequential`	from $0.030/image	Batch generation with v4.5
Seedream v4	`bytedance/seedream-v4`	from $0.028/image	Stable, cost-effective option

Prices shown are starting prices. Higher resolution or additional features may cost more.

Input Parameters

Text-to-Image (Generate Mode)

Parameter	Type	Default	Description
`prompt`	string	(required)	Text description of the desired image
`size`	string	`1024x1024`	Output image dimensions (e.g., `512x512`, `1024x1024`, `1280x720`)
`num_inference_steps`	number	`25`	Number of denoising steps (higher = better quality, slower)
`seed`	number	random	Random seed for reproducible results
`guidance_scale`	number	`5.0`	How closely to follow the prompt (higher = more faithful)
`num_images`	number	`1`	Number of images to generate (1 for standard, up to 15 for sequential)
`negative_prompt`	string	—	What to avoid in the generated image

Image Editing (Edit Mode)

Parameter	Type	Default	Description
`prompt`	string	(required)	Text description of the desired edit
`image`	string	(required)	Base64-encoded image or URL of the source image
`mask_image`	string	—	Base64-encoded mask or URL (white = edit area, black = preserve)
`strength`	number	`0.75`	Edit strength (0.0 = no change, 1.0 = full regeneration)
`size`	string	from source	Output image dimensions
`num_inference_steps`	number	`25`	Number of denoising steps
`seed`	number	random	Random seed for reproducible results
`guidance_scale`	number	`5.0`	How closely to follow the prompt
`negative_prompt`	string	—	What to avoid in the edit

Sequential / Batch Mode

Parameter	Type	Default	Description
`prompt`	string	(required)	Text description for the batch
`num_images`	number	(required)	Number of images to generate (up to 15)
`size`	string	`1024x1024`	Output image dimensions
`num_inference_steps`	number	`25`	Number of denoising steps
`seed`	number	random	Base random seed
`guidance_scale`	number	`5.0`	How closely to follow the prompt
`negative_prompt`	string	—	What to avoid in the generated images

Why Seedream?

Best Chinese Text Rendering

Seedream is the undisputed leader in rendering Chinese characters within generated images. Whether you're creating posters, banners, social media graphics, or product mockups with Chinese typography, Seedream produces crisp, accurate, and beautifully styled text that other models simply cannot match.

Poster & Design Excellence

The Seedream model family has been specifically optimized for commercial design use cases:

Magazine covers with perfect text placement
Product posters with accurate brand typography
Social media graphics with legible overlay text
Event banners with complex multi-line layouts
Marketing materials with mixed Chinese/English text

NSFW Whitelisted on Atlas Cloud

Atlas Cloud provides whitelisted access to Seedream's full capabilities, including NSFW content generation. This is ideal for:

Adult content platforms
Artistic nude photography generation
Unrestricted creative exploration
Content that other platforms would block

Sequential Generation

Need a series of related images? Seedream's sequential mode generates up to 15 coherent, thematically linked images in a single API call — perfect for:

Storyboard creation
Product angle variations
Style exploration
A/B testing visual concepts
Comic or manga panel generation

Installation

Prerequisites

Bun runtime installed
An Atlas Cloud API key (get one here)

Setup

# Clone the repository
git clone https://github.com/thoughtincode/seedream-image-skill.git
cd seedream-image-skill

# Install dependencies
bun install

# Set up your API key
cp .env.example .env
# Edit .env and add your ATLAS_CLOUD_API_KEY

# Link the CLI globally
bun link

Verify Installation

seedream --help

Usage

Basic Text-to-Image

seedream "A majestic dragon flying over a mountain range at sunset"

Specify Model and Size

seedream "Cyberpunk cityscape at night" --model v4.5 --size 1280x720

Image Editing

seedream "Add a rainbow in the sky" --mode edit --image ./photo.png --strength 0.6

Edit with Mask

seedream "Replace with a red sports car" --mode edit --image ./scene.png --mask ./mask.png

Batch Generation

seedream "Cute cat in different poses" --mode batch --count 5

Sequential with Custom Parameters

seedream "Comic panels of a space adventure" --mode batch --count 8 --size 768x768 --steps 30

NSFW Mode

seedream "Artistic figure study, oil painting style" --nsfw

Full Parameter Control

seedream "A beautiful Japanese garden in autumn" \
  --model v5.0-lite \
  --size 1024x1024 \
  --steps 30 \
  --guidance 7.5 \
  --seed 42 \
  --output ./my-garden.png \
  --negative-prompt "blurry, low quality"

CLI Reference

seedream <prompt> [options]

Arguments:
  prompt                    Text description of the image to generate

Options:
  --model <variant>         Model variant: v5.0-lite (default), v4.5, v4
  --mode <mode>             Generation mode: generate (default), edit, batch
  --size <WxH>              Image dimensions (default: 1024x1024)
  --steps <n>               Inference steps (default: 25)
  --guidance <n>            Guidance scale (default: 5.0)
  --count <n>               Number of images for batch mode (default: 1, max: 15)
  --seed <n>                Random seed for reproducibility
  --output <path>           Output file path (default: ./output/<timestamp>.png)
  --image <path>            Source image for edit mode (file path or URL)
  --mask <path>             Mask image for edit mode (file path or URL)
  --strength <n>            Edit strength 0.0-1.0 (default: 0.75)
  --negative-prompt <text>  What to avoid in the generation
  --nsfw                    Enable NSFW content generation
  --help                    Show this help message

Agent Skill Integration

This tool is designed to work as an AI agent skill across all major coding agents. Install it with a single command:

npx skills add seedream-image-skill

This works with Claude Code, Cursor, OpenAI Codex, GitHub Copilot, Gemini CLI, Windsurf, OpenCode, Kiro, and any agent that supports skill installation.

Add to your project's CLAUDE.md

## Image Generation with Seedream

Use the `seedream` CLI to generate images:
- Basic: `seedream "your prompt here"`
- Edit: `seedream "edit instructions" --mode edit --image ./source.png`
- Batch: `seedream "prompt" --mode batch --count 5`
- NSFW: add `--nsfw` flag
- Models: v5.0-lite (default), v4.5, v4

How It Works

Your AI agent reads the skill instructions
When asked to generate images, the agent invokes the seedream CLI
The CLI sends requests to Atlas Cloud API
Images are downloaded and saved to the ./output directory
The agent can reference the saved images in subsequent interactions

API Details

Authentication

Set your Atlas Cloud API key in the .env file or as an environment variable:

export ATLAS_CLOUD_API_KEY=your_api_key_here

Endpoints

Endpoint	Method	Description
`https://api.atlascloud.ai/api/v1/model/generateImage`	POST	Submit image generation request
`https://api.atlascloud.ai/api/v1/model/result/{request_id}`	GET	Poll for generation result

Request Flow

Submit — POST to /generateImage with model ID and parameters
Poll — GET /result/{request_id} until status is completed or failed
Download — Retrieve generated image(s) from the response URLs

Response Format

{
  "request_id": "abc123",
  "status": "completed",
  "output": {
    "images": [
      {
        "url": "https://...",
        "content_type": "image/png"
      }
    ]
  }
}

Output

Generated images are saved to the ./output directory by default. Each file is named with a timestamp and sequence number:

output/
  seedream-2026-03-12-143022-001.png
  seedream-2026-03-12-143022-002.png
  ...

Use --output <path> to specify a custom output location.

Examples

Chinese Typography Poster

seedream "一张现代感十足的电影海报，标题'星际迷航'用金色大字居中，背景是深邃的星空" --size 720x1280 --steps 30

Product Photography

seedream "Professional product photo of a luxury watch on marble surface, studio lighting" --guidance 7.0

Batch Style Exploration

seedream "Portrait of a woman, different artistic styles" --mode batch --count 6 --size 768x768

Edit: Background Replacement

seedream "Place the subject on a tropical beach at sunset" --mode edit --image ./portrait.png --strength 0.8

Troubleshooting

Issue	Solution
`ATLAS_CLOUD_API_KEY not set`	Set your API key in `.env` or environment
`Model not found`	Check model variant spelling
`Request timed out`	Increase poll timeout or try again
`Image too large`	Reduce dimensions or use a smaller source image
`Rate limited`	Wait a moment and retry

License

MIT License. See LICENSE for details.

Take This to Production Today

This workflow is optimized for Atlas Cloud. Move from experiment to enterprise-ready scale.

Unmatched Typography: Best Chinese text rendering in any image model
Poster Design: Purpose-built for commercial design workflows
NSFW Whitelisted: Full unrestricted generation capability
Sequential Generation: Up to 15 images per request
Enterprise Security: SOC I & II Certified | HIPAA Compliant

Start Building on Atlas Cloud

Name		Name	Last commit message	Last commit date
Latest commit History 7 Commits
scripts		scripts
src		src
.env.example		.env.example
.gitignore		.gitignore
CLAUDE.md		CLAUDE.md
LICENSE		LICENSE
README.md		README.md
SKILL.md		SKILL.md
package.json		package.json

Folders and files

Latest commit

History

Repository files navigation

Seedream Image Skill

Features

Model Variants

Input Parameters

Text-to-Image (Generate Mode)

Image Editing (Edit Mode)

Sequential / Batch Mode

Why Seedream?

Best Chinese Text Rendering

Poster & Design Excellence

NSFW Whitelisted on Atlas Cloud

Sequential Generation

Installation

Prerequisites

Setup

Verify Installation

Usage

Basic Text-to-Image

Specify Model and Size

Image Editing

Edit with Mask

Batch Generation

Sequential with Custom Parameters

NSFW Mode

Full Parameter Control

CLI Reference

Agent Skill Integration

Add to your project's CLAUDE.md

How It Works

API Details

Authentication

Endpoints

Request Flow

Response Format

Output

Examples

Chinese Typography Poster

Product Photography

Batch Style Exploration

Edit: Background Replacement

Troubleshooting

License

Take This to Production Today

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages